Do We Need Linguistics When We Have Statistics? A Comparative Analysis of the Contributions of Linguistic Cues to a Statistical Word Grouping System
Vasileios Hatzivassiloglou · The MIT Press eBooks · 1996
without each such feature, obtaining an estimate of each feature’s positive or negative contribution to We present a comparative analysis of the performance of the overall performance. By matching cases where a statistics-based system for the formation of semantic groups all system parameters are the same except for one of adjectives when various sources of linguistic knowledge feature, we assess the statistical significance of the are introduced. We identify four different types of shallow differences found. Also, a statistical model of the linguistic knowledge that are applicable to this system, and system’s performance in terms of the active feawe quantify the performance gained by incorporating each tures for each run offers a view of the contribu-such knowledge module. We perform experiments for dif-tions of features from a different angle, contrasting the significance of linguistic features (or other ferent corpus sizes and different inputs (sets of adjectives to modeled system parameters) against each other. group), collect data on the usefulness of each linguistic module, assess the statistical significance of the results, and Our analysis of the experimental results compare the contributions of the linguistic knowledge sources showed that many forms of linguistic knowledge against each other. We also assess the overall effect linguistic have a significant positive contribution to the perknowledge has in our system. Our results show that linguistic formance of the system. We attribute to the com-knowledge causes a significant increase in the performance of bined effect of the linguistic knowledge modules the system. We conclude by discussing how these positive the ability of our system to perform fine-tuned classification of adjectives into semantic classes. results can be generalized to other problems in statistical Other statistical systems that address word clas-NLP. sification problems do not emphasize the use of linguistic knowledge and do not deal with a specific word class [Brown et al., 1992], or do not